V ML jsou data kritická — kvalita a množství trénovacích dat z velké části určují výkon modelu. Princip "garbage in, garbage out" se uplatňuje výrazně: i skvělé algoritmy selžou se špatnými daty, zatímco dobrá data jsou často důležitější než volba algoritmu.
Proč na datech záleží tolik
ML models LEARN from data → the data fundamentally shapes what they learn:
→ GARBAGE IN, GARBAGE OUT → poor data → poor model (no algorithm fixes bad data)
→ good DATA is often MORE impactful than the algorithm (data > model tweaks, often)
→ models can only be as good as the data they learn from
→ data is frequently the most important factor in ML success
Kvalita dat
✓ ACCURATE/correct → wrong labels/values → the model learns wrong things
✓ RELEVANT → data representative of the real problem/distribution
✓ CLEAN → handle missing values, errors, duplicates, noise
✓ UNBIASED → biased data → biased model (perpetuates/amplifies bias — a serious issue)
✓ CONSISTENT, well-labeled → good labels are crucial for supervised learning
